[PATCH Lucid CVE-2012-3412] sfc: Fix maximum number of TSO segments and minimum TX queue size

Tim Gardner tim.gardner at canonical.com
Fri Sep 21 16:27:21 UTC 2012

From: Ben Hutchings <ben at decadent.org.uk>


BugLink: http://bugs.launchpad.net/bugs/1037456

This is related to commit 7e6d06f0de3f74ca929441add094518ae332257c
upstream, but looks very different because:

- TX queue size was constant before 2.6.37, so we don't need to check it
- The upstream fix relies on changes to the TCP stack and networking
  core, which are not appropriate for stable updates.  Instead we limit
  number of segments in efx_enqueue_skb_tso().  This effectively drops
  all the extra packets and seriously hurts TCP throughput if the limit
  is ever hit, but this shouldn't affect any legitimate traffic.

The original commit message is:

Currently an skb requiring TSO may not fit within a minimum-size TX
queue.  The TX queue selected for the skb may stall and trigger the TX
watchdog repeatedly (since the problem skb will be retried after the
TX reset).  This issue is designated as CVE-2012-3412.

Set the maximum number of TSO segments for our devices to 100.  This
should make no difference to behaviour unless the actual MSS is less
than about 700.  Increase the minimum TX queue size accordingly to
allow for 2 worst-case skbs, so that there will definitely be space
to add an skb after we wake a queue.

To avoid invalidating existing configurations, change
efx_ethtool_set_ringparam() to fix up values that are too small rather
than returning -EINVAL.

Signed-off-by: Ben Hutchings <bhutchings at solarflare.com>
Signed-off-by: Tim Gardner <tim.gardner at canonical.com>
 drivers/net/sfc/efx.h |    3 +++
 drivers/net/sfc/tx.c  |   15 +++++++++++++++
 2 files changed, 18 insertions(+)

diff --git a/drivers/net/sfc/efx.h b/drivers/net/sfc/efx.h
index 8141c76..dc0bd22 100644
--- a/drivers/net/sfc/efx.h
+++ b/drivers/net/sfc/efx.h
@@ -45,6 +45,9 @@ extern void efx_wake_queue(struct efx_nic *efx);
 #define EFX_TXQ_MASK (EFX_TXQ_SIZE - 1)
+/* Maximum number of TCP segments we support for soft-TSO */
+#define EFX_TSO_MAX_SEGS	100
 /* RX */
 extern int efx_probe_rx_queue(struct efx_rx_queue *rx_queue);
 extern void efx_remove_rx_queue(struct efx_rx_queue *rx_queue);
diff --git a/drivers/net/sfc/tx.c b/drivers/net/sfc/tx.c
index 585772b..7fdbf78 100644
--- a/drivers/net/sfc/tx.c
+++ b/drivers/net/sfc/tx.c
@@ -1072,6 +1072,21 @@ static int efx_enqueue_skb_tso(struct efx_tx_queue *tx_queue,
 	int frag_i, rc, rc2 = NETDEV_TX_OK;
 	struct tso_state state;
+	/* Since the stack does not limit the number of segments per
+	 * skb, we must do so.  Otherwise an attacker may be able to
+	 * make the TCP produce skbs that will never fit in our TX
+	 * queue, causing repeated resets.
+	 */
+	if (unlikely(skb_shinfo(skb)->gso_segs > EFX_TSO_MAX_SEGS)) {
+		unsigned int excess =
+			(skb_shinfo(skb)->gso_segs - EFX_TSO_MAX_SEGS) *
+			skb_shinfo(skb)->gso_size;
+		if (__pskb_trim(skb, skb->len - excess)) {
+			dev_kfree_skb_any(skb);
+			return NETDEV_TX_OK;
+		}
+	}
 	/* Find the packet protocol and sanity-check it */
 	state.protocol = efx_tso_check_protocol(skb);

More information about the kernel-team mailing list